Design of crossbar architecture for vector processing

Abstract

This research aims at modeling the effect of Roff to Ron ratio for a binary Resistive Random Access Memory (RRAM) based crossbar architecture with specific focus on deep learning application such as image classification. The crossbar structure uses emerging non-volatile memory (eNVM) array architecture and is simulated with complex neural networks to obtain metrics such as accuracy, inference and run-time. Model validation is performed by running benchmark image datasets. It will be possible to obtain other hardware results when this project is implemented on actual hardware.